-
Notifications
You must be signed in to change notification settings - Fork 16
Description
First off, thank you for developing this add-on! Its very useful
Reducing reviews
I believe the scheduling efficiency could be enhanced by tying a card's review interval directly to the intervals of the morphs it contains.
My suggestion is to set a card's review interval to be the minimum interval of all the morphs in its sentence.
For example, consider a card with the sentence "Hi, how are you?".
If the morph "you" has the shortest review interval among all the morphs in that sentence, the entire card would be scheduled for review based on that shorter interval.
Conversely, if all the morphs in "Hi, how are you?" have very long intervals (because we have successfully reviewed them in other sentences recently), this logic would correctly assign a long review interval to the card, preventing an unnecessary and premature review.
This would ensure that cards are only reviewed when their most-needed component ("the weakest link") is actually due, making the review process more efficient and targeted.
Dynamically creating optimal cards
Currently, we rely on a large deck in order to find an optimal 1T sentence to schedule for review. I propose using an LLM to generate 1T sentences for us. I think we could create a token mask using all known morphs in order restrict generation to known words, and then also force the target morph to appear in the result.
To make sure the generated sentence makes sense, we can check the sentences perplexity score / evaluate grammatical correctness.
Does that make sense? I'd interested in working on these things, but I'd love to get your feedback @mortii